Perceptron Mk1.5: 35B egocentric model for embodied agents
Tool-augmented egocentric video looks set to become the default recipe for wearable and robot perception, replacing attempts to bake all world knowledge into the weights. Builders of embodied systems should study how Mk1.5 chains information-gap detection with tool selection mid-task. Compared with general multimodal models trained on third-person footage, this egocentric-first approach targets a distinctly harder input distribution.
Perceptron released Mk1.5, a 35B embodied-agent model trained specifically on first-person (egocentric) video, with up to 4.
7× faster request completion than its predecessor.The model is designed for robots and smart glasses, where hands, moving cameras and occlusion make video understanding fundamentally different from standard video tasks.
On the MMSearch benchmark, Mk1.5 jumps from 18.
2 to 54.3 F1 when tool use is enabled, a 36.
1-point gain from web search, page reading, reverse-image search and zoom.The company frames the result as evidence that embodied agents should call external tools rather than encode every capability in their weights.
Perceptron released Mk1.5, a 35B embodied-agent model with up to 4.7× faster request completion.
4. First-person video is a very different problem from ordinary video understanding.
For a robot or a pair of smart glasses, hands constantly enter the frame, manipulate objects, disappear, return, and partially occlude what matters. The camera itself is moving too.
Perceptron trained Mk1.5 specifically on egocentric video, and very interestingly the model gets a 36.1-point jump on MMSearch when tools are enabled. Moves from 18.2 to 54.3 F1 when it can use web search, page reading, reverse-image search, and zoom.
This is a useful way to think about embodied intelligence too.
A robot does not need every possible capability encoded into its weights. It needs to recognize when its current information is insufficient, pick the right external capability, use it, then continue the task.
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Perceptron, Mk1.5, multimodal, embodied AI.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai).
